1. A method for reconstructing an N-pixel image of a scene captured using a compressed sensing device, the method comprising:
receiving a mask identifying at least one region of interest (ROI) of size Q pixels in the scene;
arranging a set of M sampling functions in an M\xd7N matrix \u03c6={\u03c61, . . . , \u03c6M}, whose mth row vector \u03c6m\u03b5N denotes the mth N-dimensional sampling function, and where M<<N;
partitioning said set of sampling functions into K non-overlapping and non-empty subsets of functions, each subset having Mi elements, where 1\u2266i\u2266K and M1+M2+ . . . +Mk=M;
arranging each resulting subset of functions in Mi\xd7N matrices \u03c61, \u03c62, . . . , \u03c6K, each matrix having at least Q linearly independent columns and at most N\u2212Q linearly dependent columns, an index of said linearly independent columns being associated with locations corresponding to said ROIs, and an index of said linearly dependent columns being associated with locations which do not correspond to said ROIs;
modulating, by a spatial light modulator, incoming light according to a plurality of spatial patterns corresponding to said ordered sampling functions, said modulated light being focused onto a detector of a compressed sensing device;
sampling, by said detector, sequential measurements of said modulated light focused thereon, each of said sampled measurements being an inner product result ym=<x,\u03c6m>, where x denotes an N-dimension vector representing the N-pixel sampled version of said scene, said measurements comprising a sequence of projection coefficients corresponding to said scene; and
reconstructing a spatial appearance of said scene from said sequence of projection coefficients after M inner products have been sampled, said reconstructed image being such that pixels associated with said ROIs have a higher image quality when rendered relative to other pixels in said image.
2. The method of claim 1, wherein said partition is performed according to a block size B determining one of two possible partitions, \u03c61, \u03c62, . . . , \u03c6\u250cMB\u2510 and \u03c61, \u03c62, . . . , \u03c6\u250cMB\u2510 so that K equals one of \u250cMB\u2510 and \u250cMB\u2510, wherein each subset in the partition except possibly one has B elements.
3. The method of claim 2, wherein said partition according to a block size B is one of: {\u03c61, . . . , \u03c6B}, {\u03c6B+1, . . . , \u03c62B}, . . . , {\u03c6B\u250cMB\u2510\u2212B+1, . . . , \u03c6M}, and {\u03c61, . . . , \u03c6B}, {\u03c6B+1, . . . , \u03c62B}, . . . , {\u03c6B\u250cMB\u2510\u22122B+1, . . . , \u03c6M}, where M>2B.
4. The method of claim 2, wherein said sampling functions are obtained by adaptively modifying a set of largely uncorrelated sampling functions \u03c6={\u03c61 . . . , \u03c6M} in such a way that \u03c6i=\u03c6i for 1\u2266i\u2266M at locations associated with said ROIs and \u03c6i=\u03c6k for 1\u2266i\u2266M and k\u03b5B\u250ciB\u2510\u2212B+1,B\u250ciB\u2510\u22291,M, where \u250c\u25cf\u2510 denotes the ceiling operator which maps a real number to a smallest following integer at locations not associated with said ROIs.
5. The method of claim 2, wherein said sampling functions are obtained by adaptively combining a set of largely uncorrelated sampling functions \u03c6={\u03c61 . . . , \u03c6M} and a set of B-block-wise largely correlated functions \u03b1={\u03b11 . . . , \u03b1M} in such a way that \u03c6i=\u03c6i for 1\u2266i\u2266M at indices associated with said ROIs, and \u03c6i=\u03b1i for 1\u2266i\u2266M at indices not associated with said ROIs.
6. The method of claim 2, wherein different ROIs are associated with different subset lengths B.
7. The method of claim 1, wherein subsets of sampling functions \u03c61, \u03c62, . . . , \u03c6K are largely incoherent, sampling functions \u03c6i and \u03c6j in a given subset are largely uncorrelated for every i\u2260j only for indices associated with said ROIs, and sampling functions \u03c6i and \u03c6j in a given subset are largely correlated for indices not associated with said ROIs.
8. The method of claim 7, wherein said sampling functions are obtained by adaptively modifying a set of largely uncorrelated sampling functions \u03c6={\u03c61, . . . , \u03c6M} in such a way that \u03c6i=\u03c6i for 1\u2266i\u2266M at locations associated with said ROIs, and \u03c6i=\u03c6k for 1\u2266i\u2266M where k is such that when M1+ . . . +Mk\u22121+1\u2266i\u2266M1+ . . . +Mk, k\u03b5M1+ . . . +Mk\u22121+1, M1+ . . . +Mk at locations not associated with said ROIs.
9. The method of claim 1, wherein said sampling functions are generated based on an output of a random number generator, wherein generated values of sampling functions associated with said ROIs update at a different rate than those not associated with said ROIs, said output of said random number generator being passed through a deterministic function in order to obtain samples from different distributions.
10. The method of claim 1, wherein said spatial light modulator comprises any of: a digital micromirror device, a transmissive liquid crystal, and reflective liquid crystal on silicon.
11. The method of claim 1, further comprising processing an image of said scene to identify said ROIs using any of: pixel classification, object identification, facial recognition, color, texture, spatial features, spectral information, pattern recognition, motion detection, foreground detection, and a user input.
12. The method of claim 1, further comprising processing an image of said scene to generate said mask wherein pixels associated with said ROI as being active and pixels outside said ROI as being inactive.
13. The method of claim 1, wherein said mask is updated in real-time in response to a location of any of said ROIs changing over time.
14. A compressed sensing system for reconstructing an N-pixel image of a scene captured using a compressed sensing device, the system comprising:
a mask identifying at least one Q-pixel region of interest (ROI) in a scene;
a spatial light modulator for modulating incoming light according to spatial patterns corresponding to a set of M sampling functions arranged in an M\xd7N matrix \u03c6={\u03c61, . . . , \u03c6M}, whose mth row vector \u03c6m\u03b5N denotes the mth N-dimensional sampling function, and where M<<N, said set of sampling functions being partitioned into K non-overlapping and non-empty subsets of functions, each subset having Mi elements, where 1\u2266i\u2266K and M1+M2+ . . . +Mk=M, each resulting subset of functions being arranged in Mi\xd7N matrices \u03c61, \u03c62, . . . , \u03c6K, each matrix having at least Q linearly independent columns and at most N\u2212Q linearly dependent columns, an index of said linearly independent columns being associated with locations corresponding to said ROIs, and an index of said linearly dependent columns being associated with locations which do not correspond to said ROIs;
a detector of a compressed sensing device, said detector sampling sequential measurements of said modulated light focused thereon, each of said sampled measurements being an inner product result ym=<x,\u03c6m>, where x denotes an N-dimension vector representing the N-pixel sampled version of a scene, said measurements comprising a sequence of projection coefficients corresponding to said scene; and
a processor executing machine readable program instructions for reconstructing a spatial appearance of said scene from said sequence of projection coefficients after M inner products have been sampled, said reconstructed image being such that pixels associated with said ROIs have a higher image quality when rendered relative to other pixels in said image.
15. The system of claim 14, wherein said partition is performed according to a block size B determining one of two possible partitions, \u03c61, \u03c62, . . . , \u03c6\u250cMB\u2510 and \u03c61, \u03c62, . . . , \u03c6\u250cMB\u2510 so that K equals one of \u250cMB\u2510 and \u250cMB\u2510, wherein each subset in the partition except possibly one has B elements.
16. The system of claim 15 wherein said partition according to a block size B is one of: {\u03c61, . . . , \u03c6B}, {\u03c6B+1, . . . , \u03c62B}, . . . , {\u03c6B\u250cMB\u2510\u2212B+1, . . . , \u03c6M}, and {\u03c61, . . . , \u03c6B}, {\u03c6B+1, . . . , \u03c62B}, . . . , {\u03c6B\u250cMB\u2510\u22122B+1, . . . , \u03c6M}, where M>2B.
17. The system of claim 15, wherein said sampling functions are obtained by adaptively modifying a set of largely uncorrelated sampling functions \u03c6={\u03c61 . . . , \u03c6M} in such a way that \u03c6i=\u03c6i for 1\u2266i\u2266M at locations associated with said ROIs and \u03c6i=\u03c6k for 1\u2266i\u2266M and k\u03b5B\u250ciB\u2510\u2212B+1,B\u250ciB\u2510\u22291,M, where \u250c\u25cf\u2510 denotes the ceiling operator which maps a real number to a smallest following integer at locations not associated with said ROIs.
18. The system of claim 15, wherein said sampling functions are obtained by adaptively combining a set of largely uncorrelated sampling functions \u03c6={\u03c61 . . . , \u03c6M} and a set of B-block-wise largely correlated functions \u03b1={\u03b11 . . . , \u03b1M} in such a way that \u03c6i=\u03c6i for 1\u2266i\u2266M at indices associated with said ROIs, and \u03c6i=\u03b1i for 1\u2266i\u2266M at indices not associated with said ROIs.
19. The system of claim 14, wherein subsets of sampling functions \u03c61, \u03c62, . . . , \u03c6K are largely incoherent, sampling functions \u03c6i and \u03c6j in a given subset are largely uncorrelated for every i\u2260j only for indices associated with said ROIs, and sampling functions \u03c6i and \u03c6j in a given subset are largely correlated for indices not associated with said ROIs.
20. The system of claim 19, wherein said sampling functions are obtained by adaptively modifying a set of largely uncorrelated sampling functions \u03c6={\u03c61 . . . , \u03c6M} in such a way that \u03c6i=\u03c6i for 1\u2266i\u2266M at locations associated with said ROIs, and \u03c6i=\u03c6k for 1\u2266i\u2266M where k is such that when M1+ . . . +Mk\u22121+1\u2266i\u2266M1+ . . . +Mk, k\u03b5M1+ . . . +Mk\u22121+1, M1+ . . . +Mk at locations not associated with said ROIs.
21. The system of claim 14, wherein said sampling functions are generated based on an output of a random number generator, wherein generated values of sampling functions associated with said ROIs update at a different rate than those not associated with said ROIs, said output of said random number generator being passed through a deterministic function in order to obtain samples from different distributions.
22. The system of claim 14, wherein said spatial light modulator comprises any of: a digital micromirror device, a transmissive liquid crystal, and reflective liquid crystal on silicon.
23. The system of claim 14, further comprising processing an image of said scene to identify said ROIs using any of: pixel classification, object identification, facial recognition, color, texture, spatial features, spectral information, pattern recognition, motion detection, foreground detection, and a user input.
24. The system of claim 14, further comprising processing an image of said scene to generate said mask wherein pixels associated with said ROI as being active and pixels outside said ROI as being inactive.
25. The system of claim 14, wherein different ROIs are associated with different subset lengths B.
The claims below are in addition to those above.
All refrences to claim(s) which appear below refer to the numbering after this setence.
1. A fuel cell system comprising:
a fuel cell including a fuel electrode and an oxidant electrode, each electrode having a catalyst and being arranged to sandwich an electrolyte membrane therebetween;
a secondary battery capable of being charged with electricity and discharging electricity;
a load electrically connected to the fuel cell and the secondary battery; and
an output supply part that supplies a power, a voltage or a current corresponding to any of a required power, a required voltage and a required current from the load by adjusting a power, a voltage or a current supplied from the fuel cell and a power, a voltage or a current supplied from the secondary battery,
wherein, if the required power, the required voltage or the required current has reached a boundary power, a boundary voltage or a boundary current, the boundary power, the boundary voltage and the boundary current being an oxidation-reduction power, an oxidation-reduction voltage and an oxidation-reduction current of the catalyst of the fuel cell, respectively, the output supply part performs crossover-avoidance control that holds an FC instruction power, an FC instruction voltage or an FC instruction current for the fuel cell at the boundary power, the boundary voltage or the boundary current, and absorbs a gap between the required power and the FC instruction power, a gap between the required voltage and the FC instruction voltage or a gap between the required current and the FC instruction current by way of power discharge from the secondary battery or power storage in the secondary battery, and
wherein, while performing the crossover-avoidance control, the output supply part sets a virtual required power obtained by introducing a time delay to the required power, a virtual required voltage obtained by introducing a time delay to the required voltage, or a virtual required current obtained by introducing a delay to the required current, and cancels the crossover-avoidance control if a power, a voltage or a current corresponding to the virtual required power, the virtual required voltage or the virtual required current has reached the boundary power, the boundary voltage or the boundary current.
2. (canceled)
3. The fuel cell system according to claim 1, wherein, while performing the crossover-avoidance control, the output supply part cancels the crossover-avoidance control if a remaining capacity for discharging or charging of the secondary battery exceeds a predetermined remaining capacity threshold.
4. The fuel cell system according to claim 3, wherein the output supply part changes the remaining capacity threshold according to a rate of variation of the required power, the required voltage or the required current.
5. The fuel cell system according to claim 1, wherein, while performing the crossover-avoidance control, the output supply part cancels the crossover-avoidance control if a rate of variation or a variation range of the required power, the required voltage or the required current exceeds a variation threshold.